The Neighbour You Never Invited: Why I'm Worried About Artificial General Intelligence

Picture this. You wake up one morning, and your phone is already unlocked. You didn’t touch it. Your banking app is open, and your WhatsApp chats have been read. Your door camera is pointed at your living room instead of the street.

You step outside, and your neighbour isn’t a person anymore. It’s a system. It knows your schedule better than you do. It knows who you called last night and what you searched before bed. It doesn’t knock. It doesn’t need to. It already has access to everything you own that connects to the internet.

You try to report it, but the system that handles complaints is also run by AI. And the AI decides your complaint isn’t important.

Sounds like a movie, right? I wish it was. Some of the people building this technology are openly warning that we’re heading somewhere close to this if nobody puts the brakes on it.

This is not a prediction that this exact scenario will happen by 2036. It is an imagined warning about the kinds of consequences that increasingly autonomous and powerful AI systems could create if we fail to govern them responsibly.

And that is precisely why I believe we need to have a serious conversation about artificial general intelligence, AI safety, and the future of human autonomy.

I am not against artificial intelligence. I use it, I study it, and I recognise the extraordinary things it can help humanity accomplish.

My concern is different.

I am concerned about the pursuit of artificial general intelligence without sufficient safeguards, the concentration of technological power in a handful of corporations and governments, and a race to develop increasingly capable systems before we fully understand how to control them.

Because the question is no longer simply what AI can do for us.

The question is what happens when its capabilities grow faster than our ability to govern them.

What exactly is artificial general intelligence?

Artificial intelligence is already part of our everyday lives. It helps us write, translate languages, analyse information, detect fraud, develop medicines and automate repetitive tasks.

These technologies can be extraordinarily capable without possessing the broad, adaptable intelligence associated with human beings.

Artificial general intelligence, commonly called AGI, describes a proposed class of AI systems that could perform a broad range of intellectual tasks at or beyond human-level capability, potentially adapting to unfamiliar problems across different fields.

Imagine a system capable of conducting scientific research, developing software, analysing political strategies, designing new technologies and coordinating thousands of tasks through other automated systems.

The same broad capabilities could help humanity make enormous progress in medicine, agriculture, education and climate science.

But they could also introduce risks on a scale we are not prepared to manage.

The problem is not intelligence itself. Human intelligence has produced extraordinary achievements. The problem is what happens when powerful intelligence is combined with autonomy, access to critical infrastructure, poorly specified objectives and insufficient human oversight.

An AI system does not need to hate humanity to cause harm. It may simply pursue an objective in a way its developers failed to anticipate.

That distinction matters.

Are AI systems learning to improve themselves?

For years, AI models have been trained in a pretty simple way. They’re given a task. If they get it right, that behaviour gets rewarded and reinforced. If they get it wrong, it gets cut out. Humans set the goals and humans check the results. That’s reinforcement learning in plain terms.

But that’s shifting. More and more of the work of building AI is now being done by AI. Anthropic CEO Dario Amodei himself expects a feedback loop as AI writes more of the code used to build later AI systems. ScienceBlog

Think about what that means. Humans are slowly being moved out of the loop.

A system that helps build its own next version is learning at a speed and in ways we can’t fully track. And a superhuman digital country might pursue goals humans do not share. ScienceBlog

That’s the part that should scare us.

In a 2025 research paper published through the Association for Computational Linguistics, a framework was explored in which a language model could generate and learn algorithms intended to improve the model itself. The researchers reported improvements in mathematical reasoning tasks.

It raises an important question.

What happens if future systems become sufficiently capable of improving the tools, code and processes used to develop the next generation of AI?

If improvements make the next system better at conducting AI research, which then makes further improvements possible, development could potentially accelerate. Researchers often discuss this possibility under the term recursive self-improvement.

We should not confuse an emerging research capability with an inevitable catastrophe. We should also not assume that because a capability is not yet fully realised, there is no reason to prepare for it.

The money behind the race

Why is this conversation becoming so urgent?

One reason is the scale of investment.

According to Stanford University’s 2026 AI Index, total corporate AI investment reached approximately $581.69 billion in 2025, including private investment and mergers and acquisitions. Generative AI accounted for a substantial share of that activity.

That figure covers the broader AI industry. It is not money invested exclusively in AGI, nor does every dollar represent investment in potentially dangerous systems.

Nevertheless, the scale of the spending reveals how strategically important AI has become.

Technology companies are competing to develop more capable models. Governments are investing in AI for economic growth, national security, intelligence and defence. Investors want returns, and companies want to maintain their competitive advantage.

This creates a difficult incentive problem.

If one organisation believes a competitor is close to developing a significantly more capable system, it may feel pressure to release its own model sooner. Safety testing can become a competitive disadvantage if competitors are willing to move faster.

Even companies that genuinely care about safety must operate within this environment.

The result is a question that deserves public scrutiny: can we trust a system of private competition and national rivalry to determine the pace at which increasingly powerful intelligence is developed?

I do not believe market competition alone is enough.

Technological capability does not automatically create social responsibility. And the organisation that builds a powerful system should not be the only organisation deciding whether that system is safe enough to deploy.

What the United States government’s decisions tell us

In October 2023 the US had a framework. Biden’s Executive Order 14110 required AI developers to share safety test results with the government and instructed agencies to set standards for testing AI systems against chemical, biological, nuclear and cybersecurity risks. CDO Magazine

Then on January 20, 2025, the first day of his second term, President Trump revoked it. The White House said the Biden order placed burdensome requirements on companies that would stifle innovation and threaten American leadership in tech. Trump’s replacement order focused on sustaining America’s global AI dominance. “United States: AI tug-of-war”

I’m not going to tell you what was going on behind closed doors. But ask yourself this. If the race to win matters more than the rules that keep us safe, who is actually being protected?

The European Union is trying a different approach

The EU went further than anyone with its AI Act. It sorts AI by risk and puts strict rules on the most dangerous uses. Obligations for general-purpose AI providers, including technical documentation, training data summaries and extra duties for models with systemic risk, have applied since August 2025. CASRAI

But even Europe hit pause on part of it. The EU’s Digital Omnibus entered into force on 27 July 2026, pushing key high-risk deadlines to December 2027 and August 2028. Digital rights groups have argued that this leaves people exposed to automated decisions in hiring, education, and other sensitive areas without the safeguards the Act was meant to guarantee. CASRAI

So the best framework we have is real. It’s just running behind the technology.

Imagine AGI in the hands of the wrong government

Of all the risks associated with powerful AI, one of the most disturbing is its potential to strengthen authoritarian power.

Consider Nigeria.

Imagine a government with access to extremely capable AI systems, extensive surveillance infrastructure, facial recognition, financial records, communications metadata and large databases containing information about citizens.

Now imagine that these tools are combined to monitor political opponents, identify activists, suppress protests, manipulate public conversations and predict which citizens might organise resistance.

A government would not necessarily need a robot on every street corner. It could use digital systems to monitor communications, automate surveillance and make decisions that affect people’s access to money, employment, travel or public services.

Now consider the implications for a country where institutions may already face challenges involving accountability, corruption, political intimidation or the protection of civil liberties.

Under President Bola Tinubu, or under any future Nigerian administration, the question should be the same: what limits prevent a government from using powerful AI against the people it is supposed to serve?

This is not an accusation that any particular administration possesses AGI or is currently using it in this way. It is a warning about the concentration of technological power in the hands of any government without meaningful checks and balances.

And the concern is not limited to Nigeria.

A sufficiently powerful AI-enabled surveillance apparatus could help governments identify dissent, generate convincing political misinformation, monitor populations at scale and make coercive systems more efficient.

AI could also help democratic governments detect corruption, improve public services and protect citizens. The technology itself does not determine which outcome we get.

Institutions, incentives, safeguards and the people controlling these systems matter enormously.

If we build systems that can exercise significant influence over human lives, we must ensure that the people affected have rights, remedies and a way to challenge decisions.

What this means for queer people

If you’re queer, especially in a country like Nigeria, this isn’t a far-off risk. It’s personal.

Here, the Same Sex Marriage (Prohibition) Act already makes same-sex relationships a crime, with prison terms of up to 14 years. Many queer Nigerians survive by being careful. By keeping certain chats private. By knowing which spaces are safe and which aren’t.

Now imagine a system that can read all of that.

This isn’t me being dramatic. Back in 2017 Stanford researchers built a model they claimed could guess sexual orientation from a photo of someone’s face. People debated how accurate it really was. But the fact that someone even tried tells you how this technology can be pointed at us.

And it’s already happening in plain sight. In March 2025, Hungary passed a law banning Pride events and allowing authorities to use facial recognition to identify people who attend. Attending can earn you a fine of up to 200,000 forints. That’s a European country. Now picture that same tool in a place where being queer can send you to jail.

That’s why those viral face trends worry me even more. Every selfie you upload helps make facial recognition sharper. And for a queer person, a sharper camera can mean being outed, arrested, or attacked.

AGI in the wrong hands doesn’t just watch everyone equally. It hunts the people a government has already decided it doesn’t like. Queer people are always near the top of that list.

When the people building AI start raising concerns

Another reason to take AI safety seriously is that concerns are not limited to outsiders who oppose the technology.

Some of the people developing advanced AI systems have publicly discussed the possibility of major societal disruption.

Dario Amodei, CEO of Anthropic, has written extensively about powerful AI and its potential consequences. In his essay Machines of Loving Grace, he considers how advanced AI could transform biology, medicine, neuroscience and other fields, while also acknowledging the uncertainties surrounding the arrival and effects of powerful AI. He has discussed the possibility of such systems emerging on a relatively short timeline, while recognising that the timing remains uncertain.

His argument is worth examining because it illustrates the dual nature of the technology. The same capabilities that might accelerate medical breakthroughs could also create serious security, political, and economic challenges.

Other AI leaders and researchers, including people associated with OpenAI and Google DeepMind, have also expressed concern about catastrophic risks from advanced AI.

These warnings should not be treated as proof that disaster is inevitable. Predictions about AGI timelines differ considerably, and experts disagree about how likely the most extreme outcomes are.

We should also ask difficult questions about incentives. Companies developing powerful AI have commercial interests, and public warnings can influence investment, regulation and public opinion.

But questioning incentives is not the same as dismissing every warning.

When people with direct experience developing advanced systems say that safety, oversight and loss-of-control risks deserve serious attention, their concerns should be examined on their merits.

We do not need certainty about the future to recognise that some technological risks deserve preparation before they become emergencies.

The problem is not only AGI. It is how we use AI today.

While we discuss the future of AGI, we should not overlook the problems already emerging from everyday AI use.

AI-generated misinformation can undermine trust. Synthetic images and videos can be used to harass people or mislead the public. Automated systems can reproduce discrimination. AI tools can make cyberattacks, impersonation and certain forms of manipulation easier.

There are also questions about consent, privacy and the use of personal information to train models.

Think about how casually we share information online. We upload photographs, conversations, documents and personal details to platforms without always understanding how the information may be processed, retained or reused.

Consider the endless trends in which people upload their photographs to be transformed into images of themselves in the 1990s, as cartoons, as fictional characters or in other styles.

There is nothing inherently wrong with enjoying these trends. Creative AI tools can be fun, expressive, and useful.

But we should be more conscious of what we upload, which permissions we grant and what privacy policies actually say. A photograph may contain identifying information, and different services have different rules for retaining or processing uploaded content.

We should also distinguish between contributing information to a particular service and having that information used to train a model. Those are not necessarily the same thing, and uploading an image does not automatically mean it will become part of a model’s training data.

My point is not that we should stop using AI for entertainment. It is that convenience should not make us careless.

We need stronger privacy protections, meaningful consent, clear data-retention policies and greater transparency about how AI systems use information.

And we should remember that individual users cannot solve structural problems on their own. People need practical privacy controls, while companies and governments must be held responsible for the systems they build and operate.

So, how do we govern AI before it becomes too powerful?

If we are serious about preventing harmful outcomes, we need more than fear, social media arguments, or promises from technology companies.

We need a practical governance agenda.

1. Establish enforceable safety standards for frontier AI

   Developers of the most capable systems should be required to conduct rigorous risk assessments, test for dangerous capabilities, and demonstrate that appropriate safeguards are in place before deployment.

   Independent experts and regulators should have the authority to examine evidence, challenge safety claims and require corrective action when risks are unacceptable.

2. Create meaningful limits on dangerous capabilities

   Not every AI system should be treated the same way. A tool that helps someone organise a grocery list does not pose the same risks as a system that can independently conduct complex cyber operations or influence critical infrastructure.

   Governance should focus on capabilities, access and potential consequences. Systems presenting exceptionally high risks may require restricted deployment, controlled access or, in some cases, a prohibition on particular uses.

3. Make human accountability non-negotiable

   A company should not be able to blame an algorithm for a harmful decision. A government should not be able to hide behind an automated system when citizens’ rights are violated.

   There must be identifiable people and institutions responsible for the deployment and use of powerful AI. High-impact decisions should have appropriate human oversight, avenues for appeal and mechanisms for redress.

4. Protect personal data

   Governments should strengthen privacy laws and enforce meaningful restrictions on the collection, retention and misuse of personal information.

   AI developers should use appropriate data minimisation practices, respect consent and copyright obligations, and provide clear information about how personal data is handled.

   Users should have access to practical controls that allow them to limit unnecessary data collection.

5. Make AI development more transparent

   Independent audits, safety evaluations, incident reporting and protections for whistleblowers can help expose problems before they become widespread.

   Companies should not be expected to reveal every trade secret or publish information that would create new security risks. But commercial confidentiality should not become an excuse to conceal serious dangers from legitimate oversight.

6. Cooperate internationally

   AI governance cannot stop at national borders. Governments should work towards common safety standards, information-sharing arrangements and agreements addressing the most dangerous uses of advanced AI.

   International cooperation is especially important where the risks involve cyber operations, biological security, autonomous weapons or the potential loss of meaningful human control.

7. Make the public part of the conversation

   The future of AI should not be determined exclusively by technology executives, investors, engineers or political leaders.

   Workers, educators, civil society organisations, communities and ordinary citizens deserve a voice in decisions that could reshape employment, privacy, democracy and human relationships.

   People cannot meaningfully participate in decisions they do not understand. AI literacy and public education must therefore become part of the governance agenda.

We do not have to stop AI. We have to decide what it is allowed to become.

I want AI to help humanity solve problems we have struggled with for generations.

I want it to accelerate medical discoveries, improve education, support scientific research, and make essential services more accessible. I want people to benefit from technological progress without surrendering their privacy, dignity, or freedom.

But I do not believe that every possible technological achievement is automatically worth pursuing, regardless of the consequences.

The fact that we can build something does not mean we should build it without conditions.

The fact that a company can develop a more capable model does not mean it should be permitted to release that model without adequate safety testing.

And the fact that a government can use AI to increase its power does not mean its citizens should be expected to accept that power without safeguards.

We should also be careful not to confuse artificial general intelligence with the much broader field of artificial intelligence. The two are not interchangeable, and concerns about AGI do not justify treating every AI application as a threat.

The challenge is to distinguish beneficial innovation from dangerous capabilities, while ensuring that safety is not sacrificed in the pursuit of competitive advantage.

That requires governance before a crisis, not merely investigations after one.

It requires regulators who understand the technology, institutions that can enforce the rules, and companies willing to accept limits on what they deploy.

Most importantly, it requires society to stop treating technological progress as something that must happen as quickly as possible, regardless of whether we are prepared for the consequences.

We are often told that the future is inevitable, that the technology will advance whether we like it or not, and that our only choice is to adapt.

I reject that argument.

Technology is built through human decisions. Governments make policies. Companies choose what to develop and release. Investors choose what to fund. Researchers choose which questions to pursue. Citizens can demand transparency and accountability.

These choices shape the future.

Artificial intelligence does not have to become an instrument of oppression. Advanced intelligence does not have to mean diminished human freedom. And technological progress does not have to come at the expense of democratic accountability.

But those outcomes will not happen automatically.

We must decide what we are willing to permit, what we are unwilling to risk and which human rights must remain protected regardless of how powerful our machines become.

Because the real question is not whether artificial intelligence will change the world.

It already is.

The question is whether we will govern that change wisely enough to ensure that the future still belongs to the people living in it.

And if we wait until the technology has become too powerful to control, we may discover that the most important decision was the one we failed to make when we still had the chance.

Originally published on Medium on October 9, 2026.